Engine effective torque is a very important performance parameter as well as control parameter. In this paper, a feedforward method of engine torque estimation is presented. After the engine control parameters such as injection pulse width and engine effective torque are gained through engine steady-state experiment, three one-stage statistical models that expresses the relationship between the inputs of engine control parameters and the output of effective torque are built with linear polynomial and artificial neural network algorithm respectively, using MATLAB Model-Based Calibration Toolbox. Tested with the data from engine dynamic experiment, artificial neural network model can obtain a good precision in real time engine torque estimation.